summaryrefslogtreecommitdiff
path: root/fs
diff options
context:
space:
mode:
Diffstat (limited to 'fs')
-rw-r--r--fs/io-wq.c72
1 files changed, 40 insertions, 32 deletions
diff --git a/fs/io-wq.c b/fs/io-wq.c
index 7d2ed8c7dd31..4ce83bb48021 100644
--- a/fs/io-wq.c
+++ b/fs/io-wq.c
@@ -51,6 +51,10 @@ struct io_worker {
struct completion ref_done;
+ unsigned long create_state;
+ struct callback_head create_work;
+ int create_index;
+
struct rcu_head rcu;
};
@@ -272,24 +276,18 @@ static void io_wqe_inc_running(struct io_worker *worker)
atomic_inc(&acct->nr_running);
}
-struct create_worker_data {
- struct callback_head work;
- struct io_wqe *wqe;
- int index;
-};
-
static void create_worker_cb(struct callback_head *cb)
{
- struct create_worker_data *cwd;
+ struct io_worker *worker;
struct io_wq *wq;
struct io_wqe *wqe;
struct io_wqe_acct *acct;
bool do_create = false, first = false;
- cwd = container_of(cb, struct create_worker_data, work);
- wqe = cwd->wqe;
+ worker = container_of(cb, struct io_worker, create_work);
+ wqe = worker->wqe;
wq = wqe->wq;
- acct = &wqe->acct[cwd->index];
+ acct = &wqe->acct[worker->create_index];
raw_spin_lock_irq(&wqe->lock);
if (acct->nr_workers < acct->max_workers) {
if (!acct->nr_workers)
@@ -299,33 +297,42 @@ static void create_worker_cb(struct callback_head *cb)
}
raw_spin_unlock_irq(&wqe->lock);
if (do_create) {
- create_io_worker(wq, wqe, cwd->index, first);
+ create_io_worker(wq, wqe, worker->create_index, first);
} else {
atomic_dec(&acct->nr_running);
io_worker_ref_put(wq);
}
- kfree(cwd);
+ clear_bit_unlock(0, &worker->create_state);
+ io_worker_release(worker);
}
-static void io_queue_worker_create(struct io_wqe *wqe, struct io_wqe_acct *acct)
+static void io_queue_worker_create(struct io_wqe *wqe, struct io_worker *worker,
+ struct io_wqe_acct *acct)
{
- struct create_worker_data *cwd;
struct io_wq *wq = wqe->wq;
/* raced with exit, just ignore create call */
if (test_bit(IO_WQ_BIT_EXIT, &wq->state))
goto fail;
+ if (!io_worker_get(worker))
+ goto fail;
+ /*
+ * create_state manages ownership of create_work/index. We should
+ * only need one entry per worker, as the worker going to sleep
+ * will trigger the condition, and waking will clear it once it
+ * runs the task_work.
+ */
+ if (test_bit(0, &worker->create_state) ||
+ test_and_set_bit_lock(0, &worker->create_state))
+ goto fail_release;
- cwd = kmalloc(sizeof(*cwd), GFP_ATOMIC);
- if (cwd) {
- init_task_work(&cwd->work, create_worker_cb);
- cwd->wqe = wqe;
- cwd->index = acct->index;
- if (!task_work_add(wq->task, &cwd->work, TWA_SIGNAL))
- return;
-
- kfree(cwd);
- }
+ init_task_work(&worker->create_work, create_worker_cb);
+ worker->create_index = acct->index;
+ if (!task_work_add(wq->task, &worker->create_work, TWA_SIGNAL))
+ return;
+ clear_bit_unlock(0, &worker->create_state);
+fail_release:
+ io_worker_release(worker);
fail:
atomic_dec(&acct->nr_running);
io_worker_ref_put(wq);
@@ -343,7 +350,7 @@ static void io_wqe_dec_running(struct io_worker *worker)
if (atomic_dec_and_test(&acct->nr_running) && io_wqe_run_queue(wqe)) {
atomic_inc(&acct->nr_running);
atomic_inc(&wqe->wq->worker_refs);
- io_queue_worker_create(wqe, acct);
+ io_queue_worker_create(wqe, worker, acct);
}
}
@@ -1004,12 +1011,12 @@ err_wq:
static bool io_task_work_match(struct callback_head *cb, void *data)
{
- struct create_worker_data *cwd;
+ struct io_worker *worker;
if (cb->func != create_worker_cb)
return false;
- cwd = container_of(cb, struct create_worker_data, work);
- return cwd->wqe->wq == data;
+ worker = container_of(cb, struct io_worker, create_work);
+ return worker->wqe->wq == data;
}
void io_wq_exit_start(struct io_wq *wq)
@@ -1026,12 +1033,13 @@ static void io_wq_exit_workers(struct io_wq *wq)
return;
while ((cb = task_work_cancel_match(wq->task, io_task_work_match, wq)) != NULL) {
- struct create_worker_data *cwd;
+ struct io_worker *worker;
- cwd = container_of(cb, struct create_worker_data, work);
- atomic_dec(&cwd->wqe->acct[cwd->index].nr_running);
+ worker = container_of(cb, struct io_worker, create_work);
+ atomic_dec(&worker->wqe->acct[worker->create_index].nr_running);
io_worker_ref_put(wq);
- kfree(cwd);
+ clear_bit_unlock(0, &worker->create_state);
+ io_worker_release(worker);
}
rcu_read_lock();